Structure of KAP1 tripartite motif identifies molecularinterfaces required for retroelement silencingGuido A. Stolla,b, Shun-ichiro Odaa,b, Zheng-Shan Chonga,b, Minmin Yuc, Stephen H. McLaughlind, and Yorgo Modisa,b,1

aMolecular Immunity Unit, Department of Medicine, University of Cambridge, Medical Research Council Laboratory of Molecular Biology (MRC-LMB), CB20QH Cambridge, United Kingdom; bCambridge Institute of Therapeutic Immunology & Infectious Disease (CITIID), Department of Medicine, University ofCambridge, CB2 0AW Cambridge, United Kingdom; cX-ray Crystallography Facility, MRC-LMB, CB2 0QH Cambridge, United Kingdom; and dBiophysicsFacility, MRC-LMB, CB2 0QH Cambridge, United Kingdom

Edited by Ming-Ming Zhou, Department of Pharmacological Sciences, Icahn School of Medicine at Mount Sinai, New York, NY, and accepted by EditorialBoard Member Axel T. Brunger June 17, 2019 (received for review January 25, 2019)

Transcription of transposable elements is tightly regulated to pre-vent genome damage. KRAB domain-containing zinc finger proteins(KRAB-ZFPs) and KRAB-associated protein 1 (KAP1/TRIM28) play akey role in regulating retrotransposons. KRAB-ZFPs recognize specificretrotransposon sequences and recruit KAP1, inducing the assemblyof an epigenetic silencing complex, with chromatin remodeling activitiesthat repress transcription of the targeted retrotransposon andadjacent genes. Our biophysical and structural data show that thetripartite motif (TRIM) of KAP1 forms antiparallel dimers, which furtherassemble into tetramers and higher-order oligomers in a concentration-dependent manner. Structure-based mutations in the B-box 1 domainprevent higher-order oligomerization without significant loss ofretrotransposon silencing activity, indicating that, in contrast to otherTRIM-family proteins, self-assembly is not essential for KAP1 function.The crystal structure of the KAP1 TRIM dimer identifies the KRABdomain binding site in the coiled-coil domain near the dyad. Mutationsat this site abolished KRAB binding and transcriptional silencingactivity of KAP1. This work identifies the interaction interfaces in theKAP1 TRIM responsible for self-association and KRAB binding andestablishes their role in retrotransposon silencing.

transcriptional repressor | epigenetic silencing | endogenous retrovirus |transposable element | ubiquitin E3 ligase

Retrovirus genomes that integrate into the genome ofgermline cells are inherited by future generations. These en-dogenous retroviruses (ERVs) can retain the ability to replicate bytranscriptional amplification and expression of the viral reversetranscriptase and integrase, which convert the genome tran-scripts into DNA and reintegrate it into the host genome. Thisamplifying retrotransposition mechanism has allowed ERVs, andother retroelements such as LINEs (long interspersed nuclearelements), to accumulate, accounting for more than half of thehuman genome (1). Approximately 100 human LINEs are stillreplication-competent and cause new integration events in 2–5%of the population (2).Some ERVs and other transposable elements (TEs) have

evolved to fulfill important cellular functions. TEs drive theevolution of transcriptional networks by spreading transcriptionfactor binding sites, promoters, and other regulatory elements (1, 3).TE-derived regulatory elements are particularly important inembryogenesis, when global hypomethylation promotes transcrip-tion. A significant fraction of pluripotency-associated transcriptionfactor binding sites is located in TEs (1). TEs also serve as areservoir of genes that can be coopted by the host. For example,TE-derived proteins catalyze V(D)J recombination (4) andsyncytiotrophoblast fusion in placental development (1, 5).Transcription of TEs must be tightly regulated, however, to

prevent pathogenesis. Disruption of protein coding sequences bytransposition events can cause genetic disorders such as hemophiliaand cystic fibrosis (6). TE reactivation in somatic cells is associatedwith cancer through disruption of tumor suppressor genes orenhanced transcription of oncogenes (6, 7). Accumulation ofTE-derived nucleic acids is associated with autoimmune diseases

including geographic atrophy, lupus, and Sjögren’s syndrome (2,8). A key source of retroelement repression is the family ofKrüppel-associated box zinc-finger proteins (KRAB-ZFPs) andKRAB-associated protein 1 (KAP1, also known as TRIM28 orTIF1β) (9). KRAB-ZFPs, the largest family of mammaliantranscription factors, recognize retroelements with a variable C-terminal array of zinc fingers (10, 11). The conserved N-terminalKRAB domain recruits KAP1 (9), which serves as a platform forthe assembly of a transcriptional silencing complex of repressivechromatin-modifying enzymes including SETDB1, a histoneH3K9 methyltransferase, and the nucleosome remodeling anddeacetylase (NuRD) complex (12).KAP1 is an 835-amino acid, 89-kDa protein from the tripartite

motif (TRIM) family (Fig. 1A). Residues 57–413 contain thedefining feature of the TRIM family: an RBCC motif consistingof a RING domain, 2 B-box–type zinc fingers, and a coiled-coildomain. The RING has ubiquitin E3 ligase activity (13). Thisactivity can be directed to tumor suppressors AMPK and p53 bythe MAGE proteins, which are overexpressed in human cancers

Significance

Retroviruses can integrate their DNA into the host-cell genome.Inherited retroviral DNA and other transposable elements ac-count for more than half of the human genome. Transposableelements must be tightly regulated to restrict their pro-liferation and prevent toxic gene expression. KAP1/TRIM28 isan essential regulator of transposable element transcription.We determined the crystal structure of the KAP1 TRIM. Thestructure identifies a protein–protein interaction site requiredfor recruitment of KAP1 to transposable elements. An epige-netic gene silencing assay confirms the importance of this sitefor KAP1-dependent silencing. We also show that KAP1 self-assembles in solution, but this self-assembly is not required forsilencing. Our work provides insights into KAP1-dependent si-lencing and tools for expanding our mechanistic understandingof this process.

Author contributions: G.A.S., Z.-S.C., and Y.M. designed research; G.A.S., S.-i.O., Z.-S.C.,and Y.M. performed research; G.A.S., S.-i.O., Z.-S.C., M.Y., S.H.M., and Y.M. contributednew reagents/analytic tools; G.A.S., S.-i.O., Z.-S.C., M.Y., S.H.M., and Y.M. analyzed data;and G.A.S. and Y.M. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. M.-M.Z. is a guest editor invited by theEditorial Board.

This open access article is distributed under Creative Commons Attribution License4.0 (CC BY).

Data deposition: The atomic coordinates reported in this paper have been deposited inthe Protein Data Bank (PDB), http://www.rcsb.org/ (accession code 6QAJ). The originalexperimental X-ray diffraction images were deposited in the SBGrid Data Bank, https://data.sbgrid.org/ (dataset 637).1To whom correspondence may be addressed. Email: ymodis@mrc-lmb.cam.ac.uk.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplemental.

Published online July 9, 2019.

15042–15051 | PNAS | July 23, 2019 | vol. 116 | no. 30 www.pnas.org/cgi/doi/10.1073/pnas.1901318116

Dow

nloa

ded

by g

uest

on

June

16,

202

1

http://crossmark.crossref.org/dialog/?doi=10.1073/pnas.1901318116&domain=pdfhttp://creativecommons.org/licenses/by/4.0/http://creativecommons.org/licenses/by/4.0/http://www.rcsb.org/http://www.rcsb.org/pdb/explore/explore.do?structureId=6QAJhttps://data.sbgrid.org/https://data.sbgrid.org/mailto:ymodis@mrc-lmb.cam.ac.ukhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/cgi/doi/10.1073/pnas.1901318116

(13, 14). The resulting proteasomal degradation of AMPK andp53 has been implicated in tumorigenesis (13, 14). The centralregion of KAP1 contains a PxVxL motif that recruits Hetero-chromatin Protein 1 (HP1) and is essential for transcriptionalsilencing (15). The C-terminal region of KAP1 contains a PHD-bromodomain tandem (residues 624–812). The PHD recruits theSUMO E2 ligase Ubc9 to SUMOylate several lysines in thebromodomain, thereby acting as an intramolecular SUMO E3ligase (16, 17). KAP1 SUMOylation is required for recruitmentand activation of SETDB1 and recruitment of NuRD (15–17).KAP1 has also been reported to SUMOylate IRF7 (18). TheKAP1 RING was necessary (but not sufficient) and the PHDdispensable for IRF7 SUMOylation (18). Similarly, the RING ofPML/TRIM19 is required but not sufficient for the SUMOE3 ligase activity of PML (19, 20).The KAP1 RBCC motif has been reported to form homotrimers

and bind KRAB domains with a stoichiometry of 3:1 KAP1:KRAB (21, 22). However, this is inconsistent with more recentreports that TRIM5, TRIM25, and TRIM69 form antiparal-lel dimers, a property predicted to be conserved across theTRIM family (23–26). Moreover, various TRIMs (TRIM5, PML/TRIM19, TRIM32) further assemble into tetramers and higher-order oligomers, including 2D lattices and molecular scaffolds(as seen in PML bodies), and these higher-order assemblies areimportant for their biological activities (25, 27–29). RING do-mains, including those of TRIM5α and TRIM32, form dimers,which contribute to E3 ligase activity by priming the associatedE2 ubiquitin ligase for ubiquitin transfer (28, 30, 31). The B-box1 domain of TRIM19 (29) and B-box 2 domain of KAP1 (PDBID code 2YVR) both also form dimers. Here, we use biophysicaland structural approaches to show that KAP1 forms antiparalleldimers, which further assemble into tetramers and higher-orderoligomers in a concentration-dependent manner. Point mutantsdefective in higher-order oligomerization were not significantlyimpaired in their retroelement silencing activity in a cell-basedassay, indicating that self-assembly is not essential for the silencingfunction of KAP1. In contrast, mutation of conserved residues inthe coiled-coil domain inhibited KRAB binding and transcrip-tional silencing. This work identifies the interaction interfaces inthe KAP1 RBCC motif responsible for self-association and KRABbinding and establishes their role in retrotransposon silencing.

ResultsKAP1 Forms Dimers that Self-Assemble into Higher-Order Oligomers.Each of the domains within the RBCC motif (RING, B-box 1,B-box 2, and coiled coil) has been reported to independentlyform dimers or oligomers in TRIM-family proteins (23–29). Thepresence of 2 or more domains capable of oligomerization in-dependently can lead to polymerization of TRIMs into lattices orscaffolds (25, 27, 28), but it remains unclear whether this appliesto KAP1. To assess the self-assembly potential of KAP1, wepurified the KAP1 RBCC motif and some of its constituent do-mains to determine their hydrodynamic properties. Size-exclusionchromatography coupled with multiangle light scattering (SEC-MALS) showed that the RING domain was monomeric at con-centrations ranging from 0.1 to 1.1 mM (1 to 10.5 g L−1; Fig. 1B).A fragment containing the RING and B-box 1 domains was mostlymonomeric at 33 μM (0.5 g L−1), but, as the protein concentrationwas increased to 610 μM (9.7 g L−1), the average hydrodynamicradius and apparent molecular weight of the protein increased byas much as 33% (Fig. 1C), indicating that monomers and dimers(or higher-order oligomers) were in dynamic equilibrium witheach other in solution. We conclude that B-box 1 drives assemblyof the RING-B-box 1 fragment into weakly associated dimers, witha dissociation constant in the low micromolar range.SEC-MALS data for the whole RBCC showed unambiguously

that it was dimeric at low protein concentrations (8.7 μM, 0.35 g L−1)and that its hydrodynamic radius and apparent molecular weight

increased by as much as 68% as the protein concentrationwas increased to 0.3 mM (12.4 g L−1; Fig. 1D). The isolatedKAP1 B-box 2 crystallized as a dimer with a surface area of 1,160 Å2

buried at the dimer interface (PDB ID code 2YVR). Together,these data suggest that the coiled-coil domain of KAP1 formstight homodimers, like most if not all other members of theTRIM family, and that KAP1 dimers can self-assemble throughfurther weak homotypic interactions between the B-boxes toform higher-order oligomers.To obtain a more direct and quantitative model of KAP1 self-

association, we performed sedimentation equilibrium analyticalultracentrifugation (SE-AUC) on KAP1 RBCC at concentra-tions from 1 to 200 μM. The equilibrium sedimentation profileswere consistent with a dynamic equilibrium between dimeric andoligomeric KAP1 (SI Appendix, Fig. S1), in agreement with theSEC-MALS data. The average molecular mass at the lowestconcentration was ∼80 kDa (Fig. 1E), consistent with the forma-tion of a dimer with a subnanomolar affinity. Further oligomeri-zation was evident as the concentration increased. The observedincrease in average molecular weight of KAP1 oligomers with in-creasing protein concentration at sedimentation equilibrium couldbe explained with 2 alternative models of self-association. Themodel with the best fit was an isodesmic self-association model inwhich KAP1 dimer-to-tetramer association is followed by unlimitedconsecutive additions of dimers (Fig. 1E). A simpler 4R2 → 2R4 →R8 model with dimers, tetramers, and octamers in dynamic equi-librium produced a fit of similar quality (Fig. 1E). In support ofthe isodesmic model, the weight-average fit residuals were slightlylower at the highest protein concentrations than for the dimer–tetramer–octamer model (Fig. 1E). However, the improved fit ofthe isodesmic model could stem from the greater number of pa-rameters versus the dimer–tetramer–octamer model. Both modelsyielded a dimer–tetramer dissociation constant Kd2,4 and higher-order dissociation constants (Kd4,8 and Kdiso) on the order of10 μM. Full-length KAP1 self-assembled in a similar manner, in-dicating that higher-order oligomerization is not an artifact ofisolating the RBCC domain (Fig. 1F). We conclude that KAP1forms tight dimers, which can associate into tetramers and octamersat high local concentration of KAP1. KAP1 may also form higher-order species, but our SE-AUC data cannot definitively confirm orrule out the presence of KAP1 species larger than octamers.

Crystal Structure of the KAP1 RBCC Tripartite Motif (TRIM). To un-derstand the molecular basis of KAP1 self-assembly and identifyits oligomerization interfaces, we determined the crystal struc-ture of KAP1 RBCC. Although crystals of KAP1 RBCC werereadily obtained, they initially diffracted X-rays poorly. Crystalssuitable for structure determination were obtained by fusingbacteriophage T4 lysozyme (T4L) to the N terminus of KAP1RBCC and methylating primary amines in the purified proteinbefore crystallization (Methods). Diffraction was anisotropic,with data up to 2.63 Å resolution but with overall completenessfalling below 90% at 3.9 Å resolution (SI Appendix, Table S1).The structure was determined by single anomalous dispersion(SAD) phasing using the anomalous scattering signal from thezinc atoms in the RING and B-boxes. The asymmetric unitcontained 2 molecules. The atomic model was refined at 2.9 Åresolution (SI Appendix, Fig. S2).The overall structure of the KAP1 RBCC dimer resembles a

dumbbell (Fig. 2). The coiled-coil domain forms a 16-nm-longantiparallel coiled coil that contains all of the dimer contacts.The ends of the coiled coil are capped by a B-box 2 domain. TheRING domains (residues 63–138) are bound to one side of thecoiled coil, close to but not in contact with the B-box 2 (residues204–243) from the same subunit. Unexpectedly, there was nointerpretable electron density for B-box 1 (residues 139–203),indicating that its position relative to the other domains is vari-able and does not obey the crystallographic symmetry. The T4L

Stoll et al. PNAS | July 23, 2019 | vol. 116 | no. 30 | 15043

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

June

16,

202

1

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplemental

is rigidly linked to the RING domain via a continuous fusedα-helix consisting of residues 158–162 from T4L (numbered 51–55 in the structure) and residues 56–62 from KAP1. The onlyother significant contacts between the N-terminal fusion region

and the KAP1 RBCC are through the tobacco etch virus (TEV)protease cleavage site, which precedes the T4L and, atypically, ismostly ordered in the structure (Fig. 2). The TEV cleavage sequenceis sandwiched in an extended conformation between the T4L and

Fig. 1. Self-assembly in solution of KAP1 RBCC and its subcomponent domains. (A) Domain organization of KAP1. B1, B-box 1; B2, B-box 2. (B–D) SEC-MALSdata for the RING (B), RING-B-box 1 (C), and RBCC (D). (E) SE-AUC analysis of RBCC molecular weight as a function of protein concentration. The averagemolecular weight isotherm from individual fits at different concentrations was fitted to an isodesmic self-association model (black line) yielding dissociationconstants of Kd2,4 = 9 μM [with a 1-σ (68.3%) CI of 7–11 μM] and Kdiso = 19 μM [1-σ CI, 16–23 μM] for the dimer–tetramer and isodesmic equilibria, respectively.An alternative fit to a dimer–tetramer–octamer model (gray line) yielded dissociation constants of Kd2,4 = 12.6 μM [1-σ CI, 7.4–23.5 μM] and Kd4,8 = 6 μM [1-σCI, 3–11 μM] for the dimer–tetramer and tetramer–octamer equilibria, respectively. (F) SEC-MALS of full-length KAP1.

15044 | www.pnas.org/cgi/doi/10.1073/pnas.1901318116 Stoll et al.

Dow

nloa

ded

by g

uest

on

June

16,

202

1

https://www.pnas.org/cgi/doi/10.1073/pnas.1901318116

the coiled coil, forming multiple polar and hydrophobic contactswith both domains. Although not physiologically relevant, thesecontacts appear to stabilize the crystal lattice by constraining theorientation of the T4L relative to KAP1 RBCC. The T4L alsoforms extensive crystal packing contacts, consistent with theimproved diffraction properties of the T4L-RBCC crystals versuscrystals of the RBCC alone.The coiled-coil domain forms a helical hairpin consisting of a

15-nm-long α-helix (residues 244–348) followed by a turn and ashorter partially helical segment (residues 357–405). The firsthelical segment contains the majority of the dimer contacts,mostly hydrophobic leucine zipper-type coiled-coil interactionswith the first helical segment from the other subunit. The secondsegment packs against the first to form a 4-helix bundle aroundthe 2-fold axis of the dimer, where the second segments from the2 subunits overlap (Fig. 2), and a 3-helix bundle at the distal ends

of the dimer, where the second segments do not overlap. Thecentral portion of the second segment has poor electron density,indicating a relatively high level of conformational flexibility.The coiled-coil domain is structurally most similar to the coiled-coil domain of TRIM25 (25, 26) (Rmsd 2.6 Å), which forms adimeric antiparallel coiled coil with the same fold and similarlength and curvature. TRIM5α forms a dimeric coiled coil withthe same fold and length but lower curvature (23) (Rmsd 3.9 Å),and TRIM69 forms a dimeric coiled coil with different secondarystructure (24) (Rmsd 4.0 Å).

The KAP1 RING and B-Box 2 Do Not Form Dimers in the RBCC CrystalStructure. RING domains of E3 ubiquitin ligases recruitubiquitin-conjugated E2 ligases to the substrate and primeubiquitin transfer from the E2 ligase to the substrate by stabilizingthe E2-ubiquitin intermediate in a closed state competent for transfer

Fig. 2. Crystal structure of KAP1 RBCC. (A) Domainorganization of the crystallized construct. B1, B-box1; B2, B-box 2; CC, coiled-coil; T4L, T4 lysozyme. (B)Overall structure of the RBCC homodimer. Threeviews along or perpendicular to the dyad are shown.The components are colored as in A. Zn atoms areshown as magenta spheres. (C) View of the RBCCdimer perpendicular to the dyad, with 1 subunit shownas a cartoon and the other as a surface.

Stoll et al. PNAS | July 23, 2019 | vol. 116 | no. 30 | 15045

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

June

16,

202

1

Fig. 3. Self-assembly properties and KRAB binding activity of KAP1 RBCC mutants. (A) Positions of the mutations in the RING and B-box 2 domains. Areference RBCC dimer is colored as in Fig. 2. Adjacent RBCC dimers forming crystal packing contacts are shown in gray with their residue numbers followed byan asterisk. (B) Model of a KAP1/TRIM28 B-box 1 dimer based on the TRIM19 B-box 1 dimer structure (29) with selected residues forming dimer contactsshown. An alignment of B-box 1 sequences (Right), the TRIM28 B-box 1 model, and the TRIM19 B-box 1 structure were used to identify mutations in KAP1 B-box 1 likely to disrupt dimer contacts. Residues known or predicted to participate in dimer contacts are shown in bold typeface in the sequence alignment.(C–F) SEC-MALS data for (C) RBCC with RING domain mutations, (D) RBCC with the B-box 2 mutations, (E) RING-B-box 1 with B-box 1 mutations, (F) and RBCC(black curve, WT; red curve, B-box 1 mutant A160D/T163A/E175R). (G) Model for oligomerization of KAP1 via B-box 1.

15046 | www.pnas.org/cgi/doi/10.1073/pnas.1901318116 Stoll et al.

Dow

nloa

ded

by g

uest

on

June

16,

202

1

https://www.pnas.org/cgi/doi/10.1073/pnas.1901318116

(31, 32). This stabilization has been proposed to be dependent onRING dimerization in TRIM5α, TRIM25, TRIM32, and BIRC-family E3 ligases (28, 30, 31). The most similar RING domainstructure to the KAP1 RING domain is that of TRIM32 (28)(Rmsd 1.9 Å). However, in contrast to the TRIM32 RING, whichdimerizes in solution via α-helices flanking the core RING domain(28), the KAP1 RING domains in the RBCC dimer are located onopposite ends of the coiled-coil domain and do not form anyhomotypic contacts. The KAP1 RING domains form crystal con-tacts with the coiled-coil domain (but not the RING domain) of aneighboring RBCC dimer. Similarly, the KAP1 B-box 2 domain alsodoes not form homotypic contacts in the RBCC dimer. The B-box2 domain does form crystal contacts with B-box 2 domains from2 different neighboring RBCC dimers in the crystal lattice, but thesehomotypic contacts are distinct from those formed by the isolatedB-box 2 in solution. The latter are moreover incompatible with theRBCC dimer structure, as the coiled-coil domain blocks the B-box2 surface that mediates dimerization of the isolated B-box 2.

Higher-Order Assembly of KAP1 Dimers Is Dependent on B-Box 1Interactions. Our hydrodynamic data indicate that KAP1 dimersself-assemble into tetramers, octamers, and higher-order oligomersthrough one or both of the B-boxes. To identify the sites responsiblefor higher-order oligomerization of KAP1, we designed structure-based mutations in the B-boxes aimed at disrupting potential di-mer contacts (Fig. 3). In B-box 2, a cluster of residues involved inhomotypic crystal contacts was mutated, yielding the variantN235A/A236D/K238A/D239A/F244A/L245A (Fig. 3A). Residuesin the RING domain forming crystal contacts (with the coiled-coildomain) were mutated in a second variant, V114A/Q123A/F125A/K127A (Fig. 3A). As B-box 1 was disordered in the RBCCstructure, we mutated residues predicted to be involved in B-box 1dimerization based on a structural model of the KAP1 B-box1 dimer generated from the TRIM19 B-box 1 dimer structure(PDB ID code 2MVW) (29), yielding the variant A160D/T163A/E175R (Fig. 3B). The oligomerization potential of each of thesevariants was then assessed by SEC-MALS. Mutations in the RING

and B-box 2 domains did not alter the self-assembly properties ofKAP1 RBCC (Fig. 3 C and D). The latter was unexpected, as acrystal structure of the isolated KAP1 B-box 2 was dimeric (PDBID code 2YVR). In contrast, the B-box 1 mutations abolishedoligomerization of the RING-Box-1 fragment and almost com-pletely inhibited higher-order oligomerization of KAP1 RBCCdimers (Fig. 3 E and F). We conclude that the assembly of KAP1RBCC dimers observed at high protein concentration occurs pri-marily through dimerization of B-box 1 domain (Fig. 3G).

Self-Assembly of KAP1 RBCC Dimers Is Not Required for RetroelementSilencing. Various TRIM proteins assemble into higher-orderoligomers, 2D lattices or molecular scaffolds that are importantfor physiological function (25, 27–29). To determine whetherself-assembly of KAP1 into higher-order oligomers is requiredfor its retroelement repression, we assayed the transcriptionalsilencing activities of wild-type KAP1 and the oligomerization-deficient mutant. We used reporter constructs in which sequencesfrom an SVA-D (SINE–Variable number tandem repeat–Alu,typeD) retroelement (recognized by ZNF91) or a LINE-1 retroelement(recognized by ZNF93) cloned upstream of a minimal SV40promoter strongly enhance firefly luciferase activity unless therespective KRAB-ZFP and KAP1 are both present to repressthe reporter (11). The assay was adapted for use in KAP1-knockout(KO) HEK 293T cells (33), which were cotransfected with thereporter plasmid and plasmids encoding ZNF91 or ZNF93, KAP1(WT or mutant), and Renilla luciferase under a constitutivepromoter. Firefly luciferase luminescence from the reporter wasnormalized against the cotransfected Renilla luciferase to controlfor transfection efficiency. Mutations in the HP1-binding motifof KAP1 (R487E/V488E) abolished its transcriptional activity,whereas disruption of the PHD (C651A) resulted in moderatederepression of the reporter, consistent with previous reports(16, 34). Unexpectedly, however, the A160D/T163A/E175RB-box 1 variant had a similar repression activity as WT on the SVAand LINE-1 reporters, despite being deficient in dimer–dimerassembly (Fig. 4). We therefore conclude that self-assembly of

Fig. 4. Transcriptional silencing assays with KAP1mutants. Data are presented as fold repression ofreporter luciferase luminescence in KAP1 KOHEK293T cellstransfected with a KAP1 variant. (A) SVA reporterrepression with the oligomerization-deficient B-box1 mutant (B1mut), the KRAB binding-deficient coiled-coil mutant (CCmut), an HP1-box mutant (HP1mut),and a PHD mutant (PHDmut). (B) LINE-1 reporter re-pression with the same set of mutants as in A. Datawere normalized to KAP1 KO cells transfected with anempty vector (EV). Error bars represent SEM betweenmeasurements (n = 3). Statistical significance wasassigned as follows: not significant (n.s.), P > 0.05; *,0.05 > P > 0.01. (Lower) Western blots of cell lysatesfrom WT or KAP1 KO HEK293T cells transfected witheach of the variants (WT, CC, B1, HP1, PHD) or emptyvector (−).

Stoll et al. PNAS | July 23, 2019 | vol. 116 | no. 30 | 15047

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

June

16,

202

1

KAP1 RBCC dimers into higher-order oligomers is not requiredfor KAP1-dependent transcriptional silencing of the SVA/LINE-1retroelements under the assay conditions.

Conserved Residues in the CC Domain Bind a KRAB Domain and AreRequired for Silencing. The primary function of the RBCC domainof KAP1 in silencing is to bind the KRAB domains of KRAB-ZFPs and hence recruit KAP1 to its genomic targets. KRAB:KAP1complexes were previously reported as containing 1 KRAB moleculeand 3 KAP1 molecules (21, 22). However, this seemed unlikely giventhat KAP1 is dimeric, so we decided to reexamine the compo-sition of KRAB:KAP1 complexes. A complex of KAP1 and theKRAB domain from ZNF93, a KRAB-ZFP that binds to a LINE-1element known to be silenced by KAP1 (11), was reconstitutedby coexpressing the proteins in Escherichia coli. SEC-MALSanalysis showed that KAP1 and ZNF93 KRAB formed a stable

complex, which retained the same ability to self-assemble intohigher-order oligomers as KAP1 alone (SI Appendix, Fig. S3A).The average molecular mass derived from SEC-MALS for theKAP1-KRAB complex of 226 kDa was inconsistent with a 1:3KRAB:KAP1 stoichiometry and instead suggested that the stoi-chiometry of the complex was 1:2 KRAB:KAP1 (236 kDa theoreticalmolecular weight; Fig. 5A).Having established that each RBCC dimer binds a single

KRAB domain, we reasoned that the interaction interface mustbe located on the dyad, in the central region of the KAP1 coiled-coil domain, as every other location would result in 2 equivalentbinding sites (and a 2:2 stoichiometry). Intriguingly, examinationof our KAP1 RBCC crystal structure revealed a cluster ofsolvent-exposed hydrophobic residues near the twofold axis(V293, M297, L300; Fig. 5B). Moreover, these amino acids areconserved in KAP1 but not present in other TRIMs (SI Appendix,

Fig. 5. Formation of a 2:1 KAP1:KRAB complex and identification of KRAB binding residues in the KAP1 coiled-coil domain required for silencing. (A) SEC-MALS of full-length KAP1 bound to ZNF93 MBP-KRAB. The expected molecular weights of a KAP1 dimer and for 2:1 and 2:2 KAP1:KRAB complexes areindicated with dashed lines. The total protein concentration of each analyte was 1.2 g L−1. (B) Close-up of the cluster of solvent-exposed hydrophobic residuesnear the dyad. The variant V293S/K296A/M297A/L300S (CC mutant) was generated to test for KRAB binding. (C) Pulldown KAP1-KRAB binding assay. KAP1RBCC was incubated with Twin-StrepII-MBP-ZNF93 KRAB, and the mixture was loaded on Strep-Tactin Sepharose. Bound proteins were detected by SDS/PAGE/Coomassie. (D) SPR KAP1-KRAB binding assay. MBP-KRAB was immobilized on the chip. WT or CC mutant KAP1 RBCC were flowed over the chip. Bindingkinetics of WT RBCC: kon = 3.6 ± 0.96 × 10

4 M−1 s−1; koff = 2.7 ± 0.1 × 10−4 s−1. (E) Model for binding of KAP1 dimers to KRAB-ZFPs.

15048 | www.pnas.org/cgi/doi/10.1073/pnas.1901318116 Stoll et al.

Dow

nloa

ded

by g

uest

on

June

16,

202

1

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/cgi/doi/10.1073/pnas.1901318116

Fig. S4). To determine whether this region of the coiled-coil do-main mediates KRAB binding, we designed the variant V293S/K296A/M297A/L300S (Fig. 5B). KAP1 RBCC domain bearingthese mutations failed to bind to MBP-KRAB in a pulldown assay(Fig. 5C). Other properties of KAP1 such as dimerization andhigher-order oligomerization were unaffected (SI Appendix, Fig.S3B), indicating that the mutations did not interfere with the overallfold of the RBCC domain. Notably, the thermal stability of thisvariant as assessed by differential scanning fluorimetry (DSF) wassignificantly higher than that of WT RBCC domain, further sup-porting a functional role of these residues (SI Appendix, Fig. S3C).A second variant with mutations on the dyad on the opposite faceof the coiled-coil domain, F391A/L395S/W398A, was mostly in-soluble. The lack of binding of the V293S/K296A/M297A/L300Svariant to MBP-KRAB was confirmed with surface plasmon reso-nance (SPR) measurements. Whereas WT RBCC domain boundMBP-tagged KRAB domain with high affinity (8 ± 2 nM Kd), nobinding was observed with the coiled-coil mutant (Fig. 5D and SIAppendix, Fig. S5).The effect of the coiled-coil mutations on the transcriptional

silencing activity of KAP1 was measured by using the reporterassay described here earlier. Consistent with its inability to bindKRAB, the V293S/K296A/M297A/L300S variant lost all re-pression activity, while being expressed at similar concentrationsas WT KAP1 (Fig. 4). These data indicate that KRABs bindKAP1 on the 2-fold axis of the RBCC dimer on a surface thatincludes residues V293/K296/M297/L300 and that this bindingsurface is required for KAP1 repression activity.

DiscussionInitial biophysical studies on the KAP1 RBCC domain suggestedthat it formed trimers, both in isolation and in complex with aKRAB domain (21, 22). More recent work on other TRIMs, incontrast, demonstrated that the coiled-coil domains of variousTRIMs mediate formation of antiparallel dimers and suggeststhat this property is conserved across the entire TRIM family(23–26). Our SEC-MALS and SE-AUC data establish thatKAP1 is dimeric rather than trimeric, and that dimers self-assembleinto tetramers, octamers, and possibly higher-order species at highlocal concentration of KAP1. Notably, the cooperativity of self-assembly implies that octamers are slightly more stable than tetra-mers, which could, in principle, promote the formation of largeoligomeric assemblies of KAP1 in the nucleus. This concentration-dependent self-association of KAP1 is primarily encoded by B-box1 and is relatively weak, with a dissociation constant in the lowmicromolar range. The lack of a discrete endpoint oligomeric state(e.g., tetramers) led us to hypothesize that the RBCC may be ableto self-associate into polymeric chains or molecular scaffold (Fig.3G), as observed for other TRIMs (25, 27–29). We therefore in-vestigated the possibility that KAP1 self-assembly may contribute toits transcriptional silencing function. Conceivably, formation oflarge oligomeric assemblies of KAP1 at its genomic target loci mightamplify silencing activity by increasing the number of recruitedrepressive chromatin-modifying molecules such as SETDB1.However, KAP1 mutants deficient in self-assembly repressedtranscription from SVA and LINE-1 reporter retrotransposonswith an efficiency similar to that of WT KAP1, indicating that,at least in this setting, higher-order oligomerization is not ab-solutely required for KAP1 function.The KAP1 RING has ubiquitin E3 ligase activity (13). E3 li-

gase activity is generally thought to be dependent on RING di-merization, in the context of RBCC dimerization, in TRIM5α,TRIM25, TRIM32, and BIRC-family E3 ligases (28, 30, 31).Unexpectedly, the crystal structure of KAP1 RBCC is incom-patible with RING dimerization, as the RING domains are posi-tioned on opposite ends of the RBCC dimer (Fig. 2B). Furthermore,the KAP1 RINGs do not form significant crystal contacts and donot dimerize or self-associate in solution. The absence of homo-

typic contacts between the RINGs (or B-box 2 domains) in theRBCC dimer suggests that KAP1 may be among the minority ofE3 ligases that can promote ubiquitin transfer from E2 to sub-strate without forming RING dimers, perhaps by using structuralelements from outside the core RING domain, as seen, for ex-ample, in CBL-B (32). Alternatively, the primary function of theRING may be to contribute to SUMO E3 ligase activity, consis-tent with reports that the RINGs of KAP1 (18) and PML (19, 20)are required for SUMOylation of specific substrates.The interaction between KAP1 and KRAB-ZFPs is critical for

recruitment of KAP1 to retroelements, but the stoichiometry ofKAP1–KRAB complexes has remained unclear and the locationof the KRAB binding surface on KAP1 unknown. Our SEC-MALS data show unambiguously that the KAP1 RBCC dimercan bind only a single KRAB domain. Moreover, we demon-strate that KRAB binding occurs on the dyad of the RBCC di-mer on a surface that includes residues V293/K296/M297/L300,and that this interface is required for KAP1 repression activity. Itwas previously proposed that RING domain, B-box 2, andcoiled-coil domain of KAP1 all contribute to KRAB binding (21,22). However, these results were inferred from mutations inresidues essential for maintaining the structural integrity of theKAP1 RBCC dimer that would likely cause misfolding of theprotein. Based on the distance of the B-box 2 from the KRAB-binding residues in the coiled-coil domain and the small size ofKRAB domains, direct involvement of B-box 2 in KRAB bindingcan be ruled out (Fig. 5E). Similarly, a contribution of the RING,which is also distant from the dyad, appears highly unlikely. Wenote that, as a single KRAB domain binds to the RBCC dimeron (or near) the dyad, the contacts formed by the KRAB domainwith each of the KAP1 protomers in the dimer are necessarilydifferent, lending an inherent asymmetry to the KRAB:KAP1interaction. Hence, loss of 2-fold symmetry in the KRAB:KAP1complex is a consequence of KRAB binding to the dyad.We have mapped the specific molecular features within the

KAP1 RBCC domain that are responsible for KAP1 self-assembly and KRAB binding. We show that binding of a singleKRAB domain to the dimeric antiparallel coiled-coil domain ofKAP1 is the first step in KAP1-depedendent epigenetic silencingof retrotransposon transcription. Further studies, informed bythe work reported here, are needed to complete our under-standing of how KAP1 recognizes KRAB-ZFPs, how it activateschromatin-modifying enzymes in subsequent steps of transcrip-tional silencing, and how it promotes ubiquitination of specificsubstrate proteins without RING dimerization.

MethodsExpression Vectors. Synthetic genes encoding KAP1 RING (residues 50–146),RING-B-box 1 (RB1; residues 50–200), RBCC (residues 50–413), and full-lengthKAP1 (residues 1–835; UniProt ID code Q13263) codon-optimized for E. coliwere cloned into the first multiple cloning site (MCS) of the pETDuet plasmid(Novagen), with N-terminal hexahistidine purification (His6) tag followed bya TEV protease cleavage site (ENLYFQG). The T4L-RBCC fusion construct wasconstructed by inserting a synthetic gene encoding the RBCC motif (residues56–413) of human KAP1 codon-optimized for E. coli into the first MCS ofpETDuet. The gene was preceded by sequences encoding: a His6 tag, a TEVprotease cleavage site, and bacteriophage T4 lysozyme (T4L) with the N-terminalmethionine deleted and the last 3 residues replaced by a single alanine residue.A synthetic gene encoding the KRAB domain (residues 1–71) from ZNF93(UniProt ID code P35789) codon-optimized for E. coli was expressed from thepET20 plasmid (Novagen) with N-terminal Twin-StrepII and maltose bindingprotein (MBP) affinity tags (the MBP tag was also required for protein solubility).For coexpression with KAP1, the same ZNF93 KRAB construct was subcloned intoMCS1 of pCDFDuet.

For the transcriptional silencing assay, expression constructs containingWTor mutant full-length human KAP1 preceded by a triple FLAG tag and linkersequenceMDYKDHDGDYKDHDIDYKDDDDKGSGGwere assembled in pLEXm(35) with the NEBuilder HiFi DNA Assembly Cloning Kit (New EnglandBioLabs). The following plasmids were used: firefly luciferase reporter plasmidspGL4cp-VNTR-OCT4Enh-E2 and pGL4cp-L1PA4-OCT4Enh-E2, pCAG_ZNF91_HA

Stoll et al. PNAS | July 23, 2019 | vol. 116 | no. 30 | 15049

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

June

16,

202

1

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplemental

encoding ZNF91, pCAG_ZNF93 encoding ZNF93, and pRTTK_Renilla encodingRenilla luciferase under a constitutive promoter (33).

Protein Expression and Purification. E. coli BL21 (DE3) cells (New EnglandBioLabs) were transformed with the respective expression construct, andstarter cultures were grown overnight at 30 °C in 2× TY medium. Startercultures were used to inoculate 2× TY medium, and cells were incubated at37 °C and 220 rpm to an optical density (OD600) of 0.4–0.5. For the expressionof KAP1 constructs, cultures were then supplemented with 50 μM ZnSO4, andthe temperature of the incubator was set to 18 °C. In the case of KRAB domainconstructs, the incubator temperature was lowered to 16 °C. Protein expressionwas induced at OD600 0.8 with 0.2 mM isopropyl-β-D-thiogalactopyranoside(IPTG). After 16 h, cells were harvested by centrifugation and stored at −80 °C.All subsequent steps were performed at 4 °C.

To purify the RBCC domain of KAP1, cells were resuspended in lysis buffercontaining 50 mM Tris, pH 8, 0.3 M NaCl, 20 mM imidazole, 0.5 mM TCEP,1:10,000 (vol/vol) Benzonase (Sigma), and 1× cOmplete EDTA-free proteaseinhibitors (Roche). The cells were lysed by sonication immediately after ad-dition of 1 mM phenylmethane sulfonyl fluoride (PMSF). The lysate wasclarified by centrifugation (30 min, 40,000 × g). The supernatant was appliedto a 5-mL HisTrap HP nickel-affinity column (GE Healthcare) equilibrated inwash buffer (50 mM Tris, pH 8, 0.3 M NaCl, 20 mM imidazole, 0.5 mM TCEP).The column was washed with 30 column volumes (CVs) of wash buffer be-fore elution with elution buffer (50 mM Tris, pH 8, 0.3 M NaCl, 0.25 M im-idazole, 0.5 mM TCEP). The buffer was exchanged to 50 mM Tris, pH 8, 0.3 MNaCl, 0.5 mM TCEP, and the His6 tag was removed by incubating the proteinovernight at 4 °C with 1:50 (wt/wt) TEV protease. Following a second nickel-affinity chromatography step to remove uncleaved protein and protease,the sample was purified by size-exclusion chromatography with a HiLoad(16/600) Superdex 200-pg column (GE Healthcare) preequilibrated in 20 mMHepes, pH 8, 0.5 M NaCl, 0.5 mM TCEP.

RING and RB1 constructs were purified as described earlier except that aSuperdex 75 (10/300) column (GE Healthcare) equilibrated in 20 mM Hepes,pH 8, 0.2 M NaCl, 0.5 mM TCEP was used for the final size-exclusion chro-matography step. T4L-RBCC fusion protein was purified as the RBCC domain,except that the His6 tag was not removed. Full-length KAP1 was purified asT4L-RBCC, except that a Superose 6 increase (10/300) column (GE Healthcare)preequilibrated in 20 mM Hepes, pH 8, 0.2 M NaCl, 0.5 mM TCEP was usedfor the final size-exclusion chromatography step.

To purify MBP-tagged ZNF93 KRAB domain, bacteria pellets were resus-pended in lysis buffer (50 mM Tris, pH 8, 0.15 M NaCl, 0.5 mM TCEP, 1:10,000[vol/vol] Benzonase solution [Sigma], 1× cOmplete EDTA-free protease in-hibitors [Roche]) and lysed by sonication. The lysate was clarified by centri-fugation (30 min, 40,000 × g). The supernatant was applied to a 5 mLStrepTrap column (GE Healthcare) preequilibrated in wash buffer (50 mMTris, pH 8, 0.15 M NaCl, 0.5 mM TCEP). The column was washed with 30 CVsof wash buffer before the protein was eluted with wash buffer supple-mented with 2.5 mM D-desthiobiotin and further purified by size-exclusionchromatography using a HiLoad (16/600) Superdex 200-pg column (GEHealthcare) preequilibrated in 20 mM Hepes, pH 8, 0.5 M NaCl, 0.5 mM TCEP.

KAP1:MBP-KRAB complex was purified as the isolated KRAB domain,except that lysis andwash buffer contained 0.2MNaCl and a Superose 6 increase(10/300) column was used for the final size-exclusion chromatography step.

SEC-MALS. Protein sample (100 μL) was subjected to SEC at 293 K on aSuperdex 200 (10/300) column (GE Healthcare) preequilibrated in 20 mMHepes, pH 8, 0.5 M NaCl, 0.5 mM TCEP (for KAP1 RBCC); a Superose 6 (10/300)column in 20 mM Hepes, pH 8, 0.2 M NaCl, 0.5 mM TCEP (for full-lengthKAP1); a Superose 6 (10/300) column in 20 mM Hepes, pH 8, 0.5 M NaCl,0.5 mM TCEP (for KAP1:MBP-KRAB complex); or a Superdex 75 (10/300)column in 20 mM Hepes, pH 8, 0.2 M NaCl, 0.5 mM TCEP (for RING andRB1 constructs) with a flow rate of 0.5 mL min−1. The SEC system was cou-pled to multiangle light scattering (MALS) and quasi-elastic light scattering(QELS) modules (DAWN-8+; Wyatt Technology). Protein in the eluate was alsodetected with a differential refractometer (Optilab T-rEX; Wyatt Technology)and a UV detector at 280 nm (1,260 UV; Agilent Technology). Molar masses ofpeaks in the elution profile were calculated from the light scattering andprotein concentration and quantified by using the differential refractive indexof the peak assuming a dn/dc of 0.186 with ASTRA6 (Wyatt Technology).

Sedimentation-Equilibrium Analytical Ultracentrifugation (SE-AUC). KAP1 RBCCsamples at 0.2 mM (8 g L−1) and 12 μM (0.5 g L−1) were diluted in a 1:3 seriesin 20 mM Hepes, pH 8, 0.5 M NaCl, 0.5 mM TCEP. Samples (110 μL) wereloaded in 12-mm, 6-sector cells and centrifuged at 5, 8.5, and 15 krpm at20 °C in an An50Ti rotor in an Optima XL-I analytical ultracentrifuge

(Beckmann). At each speed, comparison of several scans was used to judgewhether equilibrium had been reached. The data were analyzed in SEDPHAT13b (36). Equilibrium sedimentation distributions were fit to obtain averagemasses. An SE-AUC average mass isotherm compiled from fits to the datawas analyzed in SEDPHAT using isodesmic and dimer–tetramer–octamer olig-omerization models. The partial-specific volumes (v-bar) and solvent densityand viscosity were calculated with Sednterp (www.rasmb.org/sednterp).

X-Ray Crystallography. Before crystallization, free amines in T4L-RBCC weremethylated by incubating 15 mL of protein solution (∼1 g L−1) in 20 mMHepes, pH 8, 0.5 M NaCl with 300 μL of 1-M dimethylamine borane complex(ABC; Sigma-Aldrich) and 600 μL of 1-M formaldehyde for 2 h at 4 °C. Anadditional 300 μL of 1-M ABC and 600 μL of formaldehyde were then added.After a further 2 h at 4 °C, 150 μL of ABC was added and the sample wasincubated overnight at 4 °C. The reaction was then quenched with 1.875 mLof 1-M Tris, pH 8. The sample was supplemented with 2 mM DTT and puri-fied with a HiLoad (16/600) Superdex 200-pg column preequilibrated in20 mM Hepes, pH 8, 0.5 M NaCl, 0.5 mM TCEP (37). Crystals were grown at18 °C by sitting drop vapor diffusion. Methylated T4L-RBCC at 4.5 g L−1

(72 μM) was mixed with an equal volume of reservoir solution optimizedfrom the Index screen (Hampton Research): 15% (wt/vol) PEG 3350, 75 mMMgCl2, 0.1 M Hepes, pH 7.5. Plate-shaped crystals appeared after 2 d andwere frozen in liquid nitrogen with 33% ethylene glycol as a cryoprotectant.X-ray diffraction data were collected at 100 K at Diamond Light Source(beamline I03) and processed with autoPROC (38) and STARANISO (GlobalPhasing). The X-ray energy was tuned to 9,672 eV, corresponding to the zincL-III edge, for data collection. Phases were determined with the singleanomalous dispersion (SAD) method in PHENIX (39) using zinc as theanomalously scattering heavy atom. Comparison of electron density mapscalculated with different high-resolution cutoffs and B-factor sharpeningfactors (SI Appendix, Fig. S2) indicated that a 2.9-Å cutoff and −33-Å2

sharpening factor produced the best map. The atomic model was built withCOOT (40) and iteratively refined with REFMAC (41) and PHENIX at 2.9-Åresolution. The atomic models for T4 lysozyme (PDB ID codes 1LYD, 2LZM)and KAP1 B-box 2 (PDB ID code 2YVR) were docked into the electron den-sity, and the rest of the atomic model was built de novo using structures ofother TRIMs as guides [PDB ID code 4LTB (25), 4TN3 (23), 5FEY (28), 5NT1(26)]. SI Appendix describes data collection and refinement statistics (SIAppendix, Table S1) and sample electron density (SI Appendix, Fig. S2).Structure figures were generated with PyMOL (Schrodinger). Methylation ofT4L-RBCC at 38 sites was confirmed by mass spectrometry (SI Appendix, Fig.S6), but no dimethyl-amine groups were visible in the map.

Atomic Model of KAP1 B-Box 1. An atomic model of the KAP1 B-box 1 domainwas generated from the TRIM19 B-box 1 structure (PDB ID code 2MVW) (29)with Phyre2 (http://www.sbg.bio.ic.ac.uk/∼phyre2/). The KAP1 B-box 1 wasthen superimposed onto each protomer of the TRIM19 B-box 1 dimer togenerate a model of the KAP1 B-box 1 dimer.

Pulldown Assay. KAP1 RBCC (8 nmol) was incubated with 2 nmol of Twin-StrepII-MBP-KRAB for 45 min on ice. StrepII-tagged bait protein was thencaptured with 100 μL of Strep-Tactin Sepharose (IBA) for 1 h at 4 °C. After4 washes with 1 mL of buffer (20 mM Hepes, pH 8, 0.5 M NaCl, 0.5 mM TCEP),the beads were boiled in 100 μL of 2× SDS/PAGE loading buffer and boundproteins were analyzed by SDS/PAGE/Coomassie.

Surface Plasmon Resonance (SPR). SPR was performed with a BiacoreT200 system with dextran-coated CM5 chips (GE Healthcare). Referencecontrol and analyte CM5 chips were equilibrated in 20 mM Hepes, pH 8.0,0.5 M NaCl, 0.5 mM TCEP at 20 °C. MBP-KRAB was immobilized onto the chipsuntil a response unit value of ∼600 was reached. SPR runs were performedwith analytes injected for 120 s followed by a 900 s dissociation in 1:2 di-lution series with initial concentrations of 34 μM for WT KAP1 and 35 μM formutant KAP1. The sensor surface was regenerated after each injection cyclewith 20 mM NaOH for 30 s with a 120-s postregeneration stabilization pe-riod. Data were fitted to a biphasic kinetic model with KaleidaGraph (Syn-ergy Software) and PRISM 8 (GraphPad) to determine kon, koff, and Kd.

Transcriptional Silencing Assay. KAP1 silencing activity was measured with areporter assays in which a SINE-VNTR-Alu (SVA) type D or LINE-1 sequenceupstream of a minimal SV40 promoter enhances firefly luciferase activityunless KAP1 and the cognate KRAB-ZFP (ZNF91 and ZNF93, respectively) arepresent to repress the reporter (11). The assay was adapted for use inHEK293T cells as described previously (33). KAP1 KO 293T cells in 24-wellplates were cotransfected with 20 ng firefly luciferase reporter plasmid, 0.2 μg

15050 | www.pnas.org/cgi/doi/10.1073/pnas.1901318116 Stoll et al.

Dow

nloa

ded

by g

uest

on

June

16,

202

1

http://www.rasmb.org/sednterphttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901318116/-/DCSupplementalhttp://www.sbg.bio.ic.ac.uk/~phyre2/https://www.pnas.org/cgi/doi/10.1073/pnas.1901318116

plasmid encoding ZNF91 or ZNF93, 0.2 μg pLEXm plasmid encoding WT ormutant KAP1, and 0.4 ng plasmid encoding Renilla luciferase using 1.5 μL ofFuGENE 6 (Promega). Luciferase activity was measured 48 h posttransfection incell lysates with the Dual Luciferase assay kit (Promega) and a Pherastar FSplate reader (BMG Labtech). Replicates were performed on separate days.Firefly luciferase values were normalized to Renilla luciferase values to controlfor transfection efficiency. Statistical significance was assessed with anunpaired t test (assuming Gaussian distributions, without Welch’s correction)with PRISM 8.

Western Blotting. HEK293T cells (4 × 105) were lysed in 100 μL Passive LysisBuffer (Promega). Cell lysates (10 μL) were boiled with 2.5 μL 4× gel loadingbuffer for 5 min at 98 °C and run on a NuPAGE 4–12% Bis-Tris poly-acrylamide gel for 45 min at 200 V. Gels were blotted by using an iBlot GelTransfer Device and Transfer Stacks (ThermoFisher). Blots were blocked with5% milk in PBS solution for 1 h at room temperature and incubated withprimary antibody overnight at 4 °C. Rabbit anti-KAP1 antibody (ab10484;Abcam) was diluted 1:10,000, whereas rabbit anti-actin antibody (ab219733;Abcam) was diluted 1:2,000. Blots were incubated with fluorescent second-ary antibodies for 30 min at room temperature. Secondary anti-rabbit an-tibodies were diluted 1:10,000. Blots were imaged by using an Odyssey CLxgel scanner (LI-COR Biosciences).

Data Availability. The structure factors and atomic coordinates were de-posited in the Protein Data Bank (PDB ID code 6QAJ) (42). The original ex-perimental X-ray diffraction images were deposited in the SBGrid Data Bank(dataset 637) (43).

Note Added in Proof. Since submission of this manuscript a crystal structure ofthe KAP1 B-box 1 dimer was published (44). Mutations at the dimer interfacedisrupt the formation of KAP1 oligomers without affecting transcriptionalsilencing, consistent with the data reported herein. A preprint was alsopublished reporting that full-length KAP1 forms elongated dimers with anative asymmetry (45).

ACKNOWLEDGMENTS. KAP1 KO cells and plasmids for the transcriptionalsilencing assay were kind gifts from Helen Rowe (University College London).The plasmids were provided with permission from David Haussler (Universityof California, Santa Cruz). We thank Christopher Douse and other membersof the laboratory of Y.M. for insightful discussions. Crystallographic datawere collected on beamline I03 at Diamond Light Source (DLS). Access to DLS(proposal MX15916) was supported by the Wellcome Trust, MRC, andBiotechnology and Biological Sciences Research Council. This work wassupported by the Wellcome Trust through Senior Research Fellowship 101908/Z/13/Z (to Y.M.) and PhD Studentship 205833/Z/16/Z (to G.A.S.).

1. M. Friedli, D. Trono, The developmental control of transposable elements and theevolution of higher species. Annu. Rev. Cell Dev. Biol. 31, 429–451 (2015).

2. J. L. Goodier, Restricting retrotransposons: A review. Mob. DNA 7, 16 (2016).3. E. B. Chuong, N. C. Elde, C. Feschotte, Regulatory evolution of innate immunity

through co-option of endogenous retroviruses. Science 351, 1083–1087 (2016).4. L. Zhou et al., Transposition of hAT elements links transposable elements and V(D)J

recombination. Nature 432, 995–1001 (2004).5. A. Dupressoir, C. Lavialle, T. Heidmann, From ancestral infectious retroviruses to bona

fide cellular genes: Role of the captured syncytins in placentation. Placenta 33, 663–671 (2012).

6. D. C. Hancks, H. H. Kazazian , Jr, Roles for retrotransposon insertions in human dis-ease. Mob. DNA 7, 9 (2016).

7. B. Lamprecht et al., Derepression of an endogenous long terminal repeat activatesthe CSF1R proto-oncogene in human lymphoma. Nat. Med. 16, 571–579, 1p following579 (2010).

8. T. Hung et al., The Ro60 autoantigen binds endogenous retroelements and regulatesinflammatory gene expression. Science 350, 455–459 (2015).

9. G. Ecco, M. Imbeault, D. Trono, KRAB zinc finger proteins. Development 144, 2719–2729 (2017).

10. M. Imbeault, P. Y. Helleboid, D. Trono, KRAB zinc-finger proteins contribute to theevolution of gene regulatory networks. Nature 543, 550–554 (2017).

11. F. M. Jacobs et al., An evolutionary arms race between KRAB zinc-finger genes ZNF91/93 and SVA/L1 retrotransposons. Nature 516, 242–245 (2014).

12. C. T. Cheng, C. Y. Kuo, D. K. Ann, KAPtain in charge of multiple missions: Emergingroles of KAP1. World J. Biol. Chem. 5, 308–320 (2014).

13. J. M. Doyle, J. Gao, J. Wang, M. Yang, P. R. Potts, MAGE-RING protein complexescomprise a family of E3 ubiquitin ligases. Mol. Cell 39, 963–974 (2010).

14. C. T. Pineda et al., Degradation of AMPK by a cancer-specific ubiquitin ligase. Cell 160,715–728 (2015).

15. S. Iyengar, P. J. Farnham, KAP1 protein: An enigmatic master regulator of the ge-nome. J. Biol. Chem. 286, 26267–26276 (2011).

16. A. V. Ivanov et al., PHD domain-mediated E3 ligase activity directs intramolecularsumoylation of an adjacent bromodomain required for gene silencing. Mol. Cell 28,823–837 (2007).

17. L. Zeng et al., Structural insights into human KAP1 PHD finger-bromodomain and itsrole in gene silencing. Nat. Struct. Mol. Biol. 15, 626–633 (2008).

18. Q. Liang et al., Tripartite motif-containing protein 28 is a small ubiquitin-relatedmodifier E3 ligase and negative regulator of IFN regulatory factor 7. J. Immunol.187, 4754–4763 (2011).

19. B. B. Quimby, V. Yong-Gonzalez, T. Anan, A. V. Strunnikov, M. Dasso, The promyelocyticleukemia protein stimulates SUMO conjugation in yeast. Oncogene 25, 2999–3005(2006).

20. T. H. Shen, H. K. Lin, P. P. Scaglioni, T. M. Yung, P. P. Pandolfi, The mechanisms ofPML-nuclear body formation. Mol. Cell 24, 331–339 (2006).

21. H. Peng et al., The structurally disordered KRAB repression domain is incorporatedinto a protease resistant core upon binding to KAP-1-RBCC domain. J. Mol. Biol. 370,269–289 (2007).

22. H. Peng et al., Reconstitution of the KRAB-KAP-1 repressor complex: A model systemfor defining the molecular anatomy of RING-B box-coiled-coil domain-mediatedprotein-protein interactions. J. Mol. Biol. 295, 1139–1162 (2000).

23. D. C. Goldstone et al., Structural studies of postentry restriction factors reveal anti-parallel dimers that enable avid binding to the HIV-1 capsid lattice. Proc. Natl. Acad.Sci. U.S.A. 111, 9609–9614 (2014).

24. Y. Li et al., Structural insights into the TRIM family of ubiquitin E3 ligases. Cell Res. 24,762–765 (2014).

25. J. G. Sanchez et al., The tripartite motif coiled-coil is an elongated antiparallel hairpindimer. Proc. Natl. Acad. Sci. U.S.A. 111, 2494–2499 (2014).

26. M. G. Koliopoulos et al., Molecular mechanism of influenza A NS1-mediatedTRIM25 recognition and inhibition. Nat. Commun. 9, 1820 (2018).

27. P. Wang et al., RING tetramerization is required for nuclear body biogenesis and PMLsumoylation. Nat. Commun. 9, 1277 (2018). Correction in: Nat. Commun. 9, 1841(2018).

28. M. G. Koliopoulos, D. Esposito, E. Christodoulou, I. A. Taylor, K. Rittinger, Functionalrole of TRIM E3 ligase oligomerization and regulation of catalytic activity. EMBO J. 35,1204–1218 (2016).

29. S. Y. Huang et al., The B-box 1 dimer of human promyelocytic leukemia protein. J.Biomol. NMR 60, 275–281 (2014).

30. Z. Yudina et al., RING dimerization links higher-order assembly of TRIM5α to synthesisof K63-linked polyubiquitin. Cell Rep. 12, 788–797 (2015).

31. H. Dou, L. Buetow, G. J. Sibbet, K. Cameron, D. T. Huang, BIRC7-E2 ubiquitin conju-gate structure reveals the mechanism of ubiquitin transfer by a RING dimer. Nat.Struct. Mol. Biol. 19, 876–883 (2012).

32. H. Dou, L. Buetow, G. J. Sibbet, K. Cameron, D. T. Huang, Essentiality of a non-RINGelement in priming donor ubiquitin for catalysis by a monomeric E3. Nat. Struct. Mol.Biol. 20, 982–986 (2013).

33. L. Robbez-Masson et al., The HUSH complex cooperates with TRIM28 to repress youngretrotransposons and new genes. Genome Res. 28, 836–845 (2018).

34. S. P. Sripathy, J. Stevens, D. C. Schultz, The KAP1 corepressor functions to coordinatethe assembly of de novo HP1-demarcated microenvironments of heterochromatinrequired for KRAB zinc finger protein-mediated transcriptional repression. Mol. Cell.Biol. 26, 8623–8638 (2006).

35. A. R. Aricescu, W. Lu, E. Y. Jones, A time- and cost-efficient system for high-levelprotein production in mammalian cells. Acta Crystallogr. D Biol. Crystallogr. 62,1243–1250 (2006).

36. P. Schuck, On the analysis of protein self-association by sedimentation velocity ana-lytical ultracentrifugation. Anal. Biochem. 320, 104–124 (2003).

37. I. Rayment et al., Three-dimensional structure of myosin subfragment-1: A molecularmotor. Science 261, 50–58 (1993).

38. C. Vonrhein et al., Data processing and analysis with the autoPROC toolbox. ActaCrystallogr. D Biol. Crystallogr. 67, 293–302 (2011).

39. P. D. Adams et al., PHENIX: A comprehensive Python-based system for macromolec-ular structure solution. Acta Crystallogr. D Biol. Crystallogr. 66, 213–221 (2010).

40. P. Emsley, K. Cowtan, Coot: Model-building tools for molecular graphics. ActaCrystallogr. D Biol. Crystallogr. 60, 2126–2132 (2004).

41. G. N. Murshudov et al., REFMAC5 for the refinement of macromolecular crystalstructures. Acta Crystallogr. D Biol. Crystallogr. 67, 355–367 (2011).

42. G. A. Stoll, S.-i. Oda, M. Yu, Y. Modis, Structure of the tripartite motif of KAP1/TRIM28. Protein Data Bank. https://www.rcsb.org/structure/6qaj. Deposited 19 De-cember 2018.

43. G. A. Stoll, M. Yu, Y. Modis, X-ray diffraction data from KAP1/TRIM28. SBGrid DataBank. https://data.sbgrid.org/dataset/637/. Deposited 19 December 2018.

44. Y. Sun et al., A dissection of oligomerization by the TRIM28 tripartite motif and theinteraction with members of the Krab-ZFP family. J. Mol. Biol. 431, 2511–2527 (2019).

45. G. Fonti et al., KAP1 is an antiparallel dimer with a natively functional asymmetry.https://doi.org/10.1101/553511 (19 February 2019).

Stoll et al. PNAS | July 23, 2019 | vol. 116 | no. 30 | 15051

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

June

16,

202

1

https://www.rcsb.org/structure/6qajhttps://data.sbgrid.org/dataset/637/https://doi.org/10.1101/553511